precommit

Signed-off-by: NickLucche <[email protected]>

Author: NickLucche

vllm/lora/layers/base_linear.py

@@ -157,8 +157,12 @@ def apply(
         # In transformers backend, x and output have extra batch dimension like
         # (1, seq_len, hidden_dim), while punica expects (seq_len, hidden_dim),
         # therefore we need to flatten the batch dimensions.
-        if (x.shape[0] ==1 and x.ndim == 3 and
-            output.shape[0] == 1 and output.ndim == 3):
+        if (
+            x.shape[0] == 1
+            and x.ndim == 3
+            and output.shape[0] == 1
+            and output.ndim == 3
+        ):
             output = output.flatten(0, 1)
             x = x.flatten(0, 1)
 

vllm/model_executor/models/gemma3n.py

@@ -20,8 +20,8 @@
 
 import torch
 from torch import nn
-
 from transformers.models.gemma3n.configuration_gemma3n import Gemma3nTextConfig
+
 from vllm.attention import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig
@@ -53,8 +53,13 @@
 from vllm.v1.attention.backends.utils import KVSharingFastPrefillMetadata
 
 from .interfaces import SupportsLoRA, SupportsQuant
-from .utils import (AutoWeightsLoader, extract_layer_index,
-                    is_pp_missing_parameter, make_layers, maybe_prefix)
+from .utils import (
+    AutoWeightsLoader,
+    extract_layer_index,
+    is_pp_missing_parameter,
+    make_layers,
+    maybe_prefix,
+)
 
 logger = init_logger(__name__)
 

vllm/model_executor/models/gemma3n_mm.py

@@ -55,10 +55,19 @@
 from vllm.sequence import IntermediateTensors
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
 
-from .interfaces import (MultiModalEmbeddings, SupportsLoRA,
-                         SupportsMultiModal, SupportsTranscription)
-from .utils import (AutoWeightsLoader, WeightsMapper, flatten_bn,
-                    init_vllm_registered_model, maybe_prefix)
+from .interfaces import (
+    MultiModalEmbeddings,
+    SupportsLoRA,
+    SupportsMultiModal,
+    SupportsTranscription,
+)
+from .utils import (
+    AutoWeightsLoader,
+    WeightsMapper,
+    flatten_bn,
+    init_vllm_registered_model,
+    maybe_prefix,
+)
 
 logger = init_logger(__name__)
 
@@ -397,8 +406,11 @@ def __init__(
         self.vocab_size = multimodal_config.vocab_size
         self.text_hidden_size = text_config.hidden_size
 
-        lora_vocab = (lora_config.lora_extra_vocab_size *
-                      (lora_config.max_loras or 1)) if lora_config else 0
+        lora_vocab = (
+            (lora_config.lora_extra_vocab_size * (lora_config.max_loras or 1))
+            if lora_config
+            else 0
+        )
         self.vocab_size = self.vocab_size + lora_vocab
 
         self.embedding = VocabParallelEmbedding(
@@ -446,7 +458,8 @@ def forward(
         """  # noqa: E501
         if (input_ids is None) ^ (inputs_embeds is not None):
             raise ValueError(
-                "You must specify exactly one of input_ids or inputs_embeds")
+                "You must specify exactly one of input_ids or inputs_embeds"
+            )
         if inputs_embeds is not None:
             emb_norm = self.soft_embedding_norm(inputs_embeds)
         else:
@@ -457,11 +470,14 @@ def forward(
         return self.embedding_post_projection_norm(emb_norm_proj)
 
 
-@MULTIMODAL_REGISTRY.register_processor(Gemma3nMultiModalProcessor,
-                                        info=Gemma3nProcessingInfo,
-                                        dummy_inputs=Gemma3nDummyInputsBuilder)
-class Gemma3nForConditionalGeneration(nn.Module, SupportsMultiModal,
-                                      SupportsTranscription, SupportsLoRA):
+@MULTIMODAL_REGISTRY.register_processor(
+    Gemma3nMultiModalProcessor,
+    info=Gemma3nProcessingInfo,
+    dummy_inputs=Gemma3nDummyInputsBuilder,
+)
+class Gemma3nForConditionalGeneration(
+    nn.Module, SupportsMultiModal, SupportsTranscription, SupportsLoRA
+):
     merge_by_field_config = True
     supported_languages = ISO639_1_SUPPORTED_LANGS
 
@@ -504,12 +520,12 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
 
         self.vision_tower = AutoModel.from_config(config=config.vision_config)
         self.audio_tower = AutoModel.from_config(config=config.audio_config)
-        self.embed_vision = Gemma3nMultimodalEmbedder(config.vision_config,
-                                                      config.text_config,
-                                                      self.lora_config)
-        self.embed_audio = Gemma3nMultimodalEmbedder(config.audio_config,
-                                                     config.text_config,
-                                                     self.lora_config)
+        self.embed_vision = Gemma3nMultimodalEmbedder(
+            config.vision_config, config.text_config, self.lora_config
+        )
+        self.embed_audio = Gemma3nMultimodalEmbedder(
+            config.audio_config, config.text_config, self.lora_config
+        )
 
         self.language_model: nn.Module = init_vllm_registered_model(
             vllm_config=vllm_config,
@@ -744,7 +760,8 @@ def get_mm_mapping(self) -> MultiModelKeys:
         return MultiModelKeys.from_string_field(
             language_model="language_model",
             connector="multi_modal_projector",
-            tower_model=["vision_tower", "audio_tower"])
+            tower_model=["vision_tower", "audio_tower"],
+        )
 
     @classmethod
     def get_placeholder_str(cls, modality: str, i: int) -> Optional[str]: